A real-time angle aware face recognition system based on artificial neural network

The increased need of person verification in daily life created a big market for biometric machine authentication tools. A few years back, fingerprint verification was done only in criminal investigation. Now finger-prints or face-images are widely used in bank tellers, airports, building entrances. Due to natural inhibition to allow finger-prints, and physical difficulties to procure it, its popularity as a biometric information can not be widely used. Face-image, on the other hand, is easy to obtain even from a distance. But its success greatly depends proper orientation and illumination of the subject´s face image, compared to that taken at the registration time. Facial features heavily change with face orientation angle - leading to increased false-rejection as well as false-acceptance. Registering face images for all possible angles is almost impossible. Our motivation is to build an angle-orientation aware face recognition technique. In this work, we proposed an memory-efficient way to register (store) multiple angle face-image data, and a computational-efficient authentication technique, using multi-layer perceptron (MLP). We use angle-features as input to the MLP, and facial-image features (using PCA and ICA) as output of the MLP. Proper angle features were selected by extensive experiments. With the available face images, we could achieve a zero equal error rate (EER) could be achieved.